Phylogenetic and Structural Analysis of Hydra ADAR

Abstract

Adenosine deaminases acting on RNAs (ADARs) perform adenosine-to-inosine (A-to-I) RNA editing for essential biological functions. While studies of editing sites in diverse animals have revealed unique biological roles of ADAR editing including temperature adaptation and reproductive maturation, rigorous biochemical and structural studies of these ADARs are lacking. Here, we present a phylogenetic sequence analysis and AlphaFold computational structure prediction to reveal that medusozoan ADAR2s contain five dsRNA binding domains (dsRBDs) with several RNA binding residues in the dsRBDs and deaminase domain conserved. Additionally, we identified evolutionary divergence between the medusozoan (e.g. Hydra) and anthozoan cnidarian subphyla. The anthozoan ADAR deaminase domains more closely resembling human ADARs with longer 5’ RNA binding loops, glutamate base-flipping residues, and a conserved TWDG dimerization motif. Conversely, medusozoan ADAR deaminase domains have short 5’ binding loops, glutamine flipping residues, and non-conserved helix dimerization motif. We also report the direct detection of A-to-I RNA editing by an ADAR ortholog from the freshwater cnidarian Hydra vulgaris (hyADAR). We solved the crystal structure of the monomeric deaminase domain of hyADAR (hyADARd) to 2.0 Å resolution, showing conserved active site architecture and the presence of a buried inositol hexakisphosphate known to be required for ADAR activity. In addition, these data demonstrate that medusozoans have evolved novel ADAR structural features, however the physiological consequence of this remains unknown. In addition, these results provide a framework for biochemically and structurally characterizing ADARs from evolutionarily distant organisms to understand the diverse roles of ADAR editing amongst metazoans.

Graphical Abstract

Introduction

While DNA has long been thought to be the determinate instructions for the generation of protein molecular machinery, modifications at the transcriptomic level have the potential to create vastly different biological outcomes. Defined as RNA editing [1, 2], these insertions, deletions, or nucleobase modifications provide a source of regulation and diversity, thus expanding the proteome beyond DNA encoded genes. One of the most common forms of RNA editing is adenosine to inosine (A-to-I) found in all domains of life [3]. A-to-I RNA editing in metazoans is carried out by Adenosine Deaminases Acting on RNA (ADARs), which catalyze the hydrolytic deamination of adenosine to inosine in duplex RNA [4, 5]. Inosine is then read by downstream cellular machinery as guanosine, resulting in an A-to-G information change at the RNA level.

ADAR editing is essential for several cellular processes, therefore aberrant editing is implicated in several human diseases [6]. ADARs play an important role in innate immunity by preventing immune activation by endogenous cytoplasmic duplex RNA [7]. In addition to innate immune regulation, ADAR editing can impact gene expression by altering miRNA biogenesis or editing 3’ untranslated regions (UTRs) of mRNAs [8]. Within coding regions of mRNAs, ADAR editing can generate proteomic diversity by deleting or introducing splice sites or changing amino acid codons [9]. Many recoded transcripts are important for neurotransmission and brain function, with a unique spatiotemporal distribution of edited transcripts occurring throughout neural development [10, 11]. One well studied example is the Q/R recoding site on the GluR-B subunit of the α-amino-2-hydroxy-5-methyl-isoxazole-4-propionate (AMPA) receptor where ADAR editing results in a glutamine to arginine amino acid change. In the unedited form, the neutral amino acid glutamine allows for calcium flux while the cationic residue arginine reduces calcium permeability [12, 13]. While these editing events have been extensively studied in the context of human ADARs, our understanding of the roles of editing in other organisms is limited.

ADAR proteins have a modular structure with a C-terminal deaminase domain and double-stranded RNA binding domains (dsRBDs) [14]. Structural studies of ADARs have been primarily focused on the human isoform ADAR2, providing insights into the base flipping mechanism and substrate specificity [15-17]. The first high-resolution structure of the deaminase domain of human ADAR2 bound to duplex RNA revealed that the base flipping mechanism of ADARs is mediated by a loop of the deaminase domain that approaches the RNA minor groove and intercalates into the duplex, flipping the target adenosine into the enzyme active site [15]. In human ADARs, as well as all other characterized ADAR homologs, this loop contains a glutamate residue (E1008 and E488 in human ADAR1 and ADAR2, respectively) that hydrogen bonds with the base that was paired with the flipped-out adenosine, or orphan base, stabilizing the base-flipped conformation. Early biochemical studies of ADAR showed that mutation of this glutamate to glutamine (E1008Q and E488Q in human ADAR1 and ADAR2, respectively) resulted increased deamination rates up to 60-fold [18]. The structure of this hyperactive ADAR2 E488Q mutant strongly resembled the structure of the wild type enzyme and suggested that the E488-orphan base interaction was protonation dependent [15]. Indeed, subsequent studies showed that pairing the wild type enzyme with proton-donor nucleobase analogs recapitulated the increased deamination rates observed with the glutamine mutant [19].

Previous structural studies of human ADARs have resulted in advances in the development of therapeutic site-directed RNA editing [19-21]. Similarly, we expect structural studies of non-human ADAR homologs will provide a more comprehensive understanding of ADAR mechanism and function. To that end, we describe the discovery of an ADAR homolog with a base-flipping loop bearing a glutamine flipping residue (synonymous with the E488Q hyperactive ADAR2 mutant) from Hydra vulgaris, a fresh-water polyp from the cnidarian phylum capable of whole-body regeneration [22]. We demonstrate this homolog contains an active deaminase domain and is predicted to contain five double stranded RNA binding domains (dsRBD), whose AlphaFold predicted structures strikingly resemble dsRBDs from human ADARs. Lastly, we solved the high-resolution crystal structure of the deaminase domain of Hydra vulgaris, the first non-human ADAR structure. We showed this structure has high similarity to structures of human ADAR2 with several conserved features resolved, including the active site architecture and buried inositol hexakisphosphate. Lastly, we identified a unique dimeric configuration of Hydra vulgaris ADAR that has not been observed in human ADARs. Taken together, these studies suggest that these structurally conserved features were likely present in a common metazoan ancestor.

Results

Phylogenetic tree of Cnidarian ADARs.

To examine the evolutionary relationship between ADARs within and beyond the cnidarian phylum, we constructed an ADAR gene tree. In this tree, the tips represent present-day gene sequences, while the nodes denote ancestral genes from which all descendant genes at the child tips have evolved. . All tips that have descended from a single common ancestral gene form an evolutionary group known as a ‘clade’. Within the cnidarian phylum are two subphyla: Medusozoa (including jellyfish and hydrozoa) and Anthozoa (including corals and sea anemones). In the ADAR gene tree, medusozoan ADAR2 genes form a single, well-supported clade that robustly clusters with all other ADARs (Figure 1a). Anthozoans possess two distinct clades of ADAR2 genes (Clades 1 and 2, Figure 1a), where Clade 1 is more closely related to the non-cnidarian ADAR2s, while Clade 2 genes are closely related to the ADAR1 clade. The ADAR1 clade is nested within the larger ADAR2 gene family (Figure 1a).

Figure 1: A maximum likelihood gene tree reveals evolutionary relationships between ADAR gene sequences.

(a)The tips of this evolutionary tree represent gene sequences. Tip names indicate species, sequence accession number, and gene name. Bootstrap scores, a measure of topological robustness, are given at the nodes of the gene tree. Bootstrap scores may range from 0 to 100, and scores of at least 95 indicate that the topology is highly robust. The tree was rooted using the ADAT1 gene family as an outgroup (ADAT1 has been collapsed for visualization, grey triangle). hyADARd is indicated with an arrow. Thick vertical lines indicate the position of ADAR1 and ADAR2 sequences. The position of medusozoan and anthozoan ADAR clades are labelled. The anthozoan ADAR2 group is paraphyletic and consists of two separate clades (Clade 1 and Clade 2). Nodes representing the last common ancestor of clades of interest are annotated with a black circle. ADAR sequences from higher eukaryotes are not boxed. (b) The most common domain architecture is given for each cnidarian clade highlighted by a colored box. Box color corresponds to domain composition: green, ADAR2 with 2-3 dsRBDS; purple, ADAR2 with 4-5 dsRBDs; blue, ADAR1 with 1-3 Z-alpha domains and 1 dsRBD.

Medusozoan ADAR2 genes robustly cluster with all other ADARs, confirming they are members of the ADAR family descended from a shared ancestral gene that gave rise to all ADARs. However, medusozoan ADAR2s form a single exclusive robustly-supported clade that diverges from all other ADARs. This likely reflects their highly unique sequence composition and suggests that medusozoan ADAR2s have undergone an independent evolutionary trajectory. Two distinct anthozoan ADAR2 clades (Clade 1 and 2) are also present, suggesting that the ADAR2 gene duplicated and diverged early in anthozoan evolution. Finally, the ADAR1 clade is wholly nested within ADAR2 sequences, suggesting that ADAR1 evolved from an ADAR2-like ancestor.

The tree sequence clustering also correlates well with conserved domains as identified by NCBI [23] and AlphaFold 3 structure predictions [24]. The medusozoan ADAR2s are all predicted to have five dsRBDs, which is unusually high compared to other organisms (Figure 1b). It is also interesting to note that the two anthozoan ADAR2 clades have distinct architectures, with Clade 1 containing the typical two dsRBDs (although E. diaphana has three), while Clade 2, like the medusozoan, contains five dsRBDs, except for one S. callimorphus sequence (v1.43643.1), which contains four dsRBDs.

Another intriguing observation of this phylogenetic tree is that Hydra does not contain any ADAR1 candidates, which contain Zα domains that binds Z-DNA and Z-RNA structures and a second non-catalytic zinc binding site. However, other Cnidarian ADAR1s are identified both among the Medusozoa and Anthozoa, again with unique domain architectures. The medusozoan ADAR1s contain only one Zα domain and one dsRBD, except R. esculentum which has two Zα domains and one dsRBD. The anthozoan ADAR1s contain either two (N. vectensis & E. diaphana) or three (S. callimorphus & P. lutea) Zα domains and one dsRBD. This ADAR1 architecture deviates from higher eukaryotes, which typically have two Zα domains and three dsRBDs. It is also interesting to note that all the predicted AlphaFold cnidarian ADAR1 structures conserve the residues (2Cys-2His) that ligate the non-catalytic second zinc site, which has been shown to impact RNA editing activity [25].

Hydra ADAR domain architecture and evolutionary conservation.

Like all ADARs, the Hydra ADAR ortholog contains a modular domain architecture with a C-terminal deaminase domain, double-stranded RNA binding domains, and a nuclear localization sequence. The number of dsRBDs varies amongst human ADARs with ADAR1 having three dsRBDs and ADAR2 and ADAR3 only have two. Interestingly, AlphaFold3 predicts the N-terminal sequence of hyADAR folds into five dsRBDs [24]. Structural alignment of each of the Hydra dsRBDs with dsRBD2 from human ADAR2 (PDB ID: 6VFF) [16] shows close similarity between the dsRBD2 of human ADAR2 and each of the predicted five Hydra vulgaris dsRBDs with several residues known to be important for RNA binding conserved (Figure 2a-f).

Figure 2: AlphaFold predicted structures of Hydra vulgaris dsRBDs align well with dsRBD from human ADAR2.

(a) Hydra vulgaris dsRBD1, (b) Hydra vulgaris dsRBD2, (c) Hydra vulgaris dsRBD3, (d) Hydra vulgaris dsRBD4, (e) Hydra vulgaris dsRBD5, (f) All five predicted dsRBDs aligned with dsRBD2 from human ADAR2 (PDB ID: 6VFF). Residues predicted to interact with RNA are shown in sticks. (g) Sequence alignment of the five dsRBDs from hydra ADAR together with the two dsRBDs of human ADAR2. Residues colored by Clustal X colored scheme mark highly conserved residues with structural and functional implications.

The dsRBD fold consists of a mixed α/β fold with a conserved α1-β1-β2-β3-α2 topology (Figure 2). Helix α2 packs against the three-stranded antiparallel β-sheet and helix α1 packs against helix α2. Helix α1 is shorter and abuts the minor groove of the dsRNA but usually does not make any sequence-specific contacts. Helix α1 typically contains a conserved glutamate residue that contacts the RNA. Based on sequence alignments of the dsRBDs (Figure 2g), this glutamate is conserved in dsRBDs 1 (E37), 2 (E163), and 4 (E334), while in dsRBD 3 this residue is N254 and in dsRBD 5 it is K424. Additionally, helix α1 is typically initiated with a proline residue, which is conserved in all five dsRBDs of hyADAR. DsRBDs also typically possess two conserved phenylalanines in β2 and β3, which is required to maintain the hydrophobic core of the dsRBD. These ten phenylalanines are conserved in all five Hydra dsRBDs, apart from two, which are replaced by the equivalent tyrosine. All five Hydra dsRBDs β2-β3 loops also contain the conserved glycine residue to preserve the type II β-hairpin turn, although in dsRBD5 the glycine is in the “i+1” position, instead of the standard “i+2” position. Finally, dsRBDs usually contain a KKxAK motif in helix α2 where the conserved lysines interact with the dsRNA phosphodiester backbone [26]. This motif is strictly conserved in Hydra dsRBDs 2, 4, and 5, while in dsRBD 1 and 3 the motif is mostly maintained with a KxxAK motif (Figure 1g). These conserved residues and predicted folds strongly suggest that Hydra ADAR contains five dsRBDs that are capable of interacting with dsRNA substrates for editing.

The deaminase domain of hyADAR most closely resembles human ADAR2, sharing 42% amino acid identity and conservation of amino acids coordinating the catalytic zinc ion and glutamate, residues responsible for binding the inositol hexakisphosphate, as well as many RNA interacting residues. Of the 15 residues identified to interact with RNA in human ADARs, 14 are completely or functionally conserved in hyADAR [16]. Significant sequence divergence between human ADAR2 and hyADAR is observed for the 5’ binding loop, which interacts with the dsRNA in the region 5’ to the target adenosine. In human ADARs, the 5’ binding loop contributes to the different substrate specificity observed between ADAR1 and ADAR2, where ADAR1 preferentially edits RNAs with a short hairpin stem and ADAR2 edits RNAs with a long duplex 5’ to the target adenosine [27-29]. One study generated an ADAR1 chimera with ADAR2 substrate specificity by grafting the 5’ binding loop of ADAR2 (residues 454-479) on ADAR1 [30]. The 5’ binding loop of hyADAR does not resemble the loop from human ADARs or ADARs from other model organisms, generally being 6-7 residues shorter (Figure 3a). Another unique feature of hyADAR falls within the base flipping loop of the protein. The flipping loop of ADARs intercalates into the RNA duplex and flips the target adenosine out of the duplex into the enzyme active site. A highly conserved glutamate on the flipping loop (E1008 and E488 in human ADAR1 and ADAR2, respectively) then hydrogen bonds with the orphan base, typically cytidine [15]. Mutation of this conserved glutamate to glutamine results in an enzyme with increased base-flipping activity and up to 60-fold higher deaminase activity on some substrates [18, 31]. To our knowledge, Hydra vulgaris ADAR is the first organism identified to have glutamine as the wild type base flipping residue, suggesting that hyADAR may have robust deaminase activity. Amino acid alignment of Hydra vulgaris ADAR with other cnidarian ADARs revealed a correlation between the length of the 5’ binding loop and the identity of the base flipping residue. Cnidarians belonging to the medusozoan subphylum have 5’ binding loops ranging from 17 to 22 residues and a glutamine flipping residue whereas cnidarians belonging to the anthozoan subphylum have longer 5’ binding loops (generally 27 residues) and a glutamate flipping residue (Figure 3b). This suggests that the medusozoan clade has evolved unique ADAR structural features that may contribute to editing activity.

Figure 3: Short 5’ binding loops of medusozoan cnidarians correlates with a glutamine flipping residue.

(a) Compared to humans and other model organisms, hyADAR has a short 5’ binding loop and a glutamine flipping residue. (b) Other cnidarians in the medusozoan subphylum also have short 5’ binding loops and glutamine flipping residues where the anthozoan subphylum has longer 5’ binding loops and glutamate flipping residues. Residues colored by Clustal X scheme in the 5’ binding loop indicated amino acids known to interact with RNA. The flipping residue is also colored by Clustal X color scheme.

Hydra vulgaris ADAR is an active deaminase.

While some studies have documented potential RNA editing sites in cnidarians including Hydra vulgaris [3], the deaminase activity of hyADAR has not been thoroughly investigated in vitro. Given that no Hydra vulgaris ADAR substrates have been validated, the deaminase activity of full-length and deaminase domain of hyADAR was tested on the well characterized human Gli1 mRNA substrate, derived from the Glioma-associated oncogene 1 mRNA under single-turnover kinetic conditions (Figure 4a) [32]. While A-to-I editing was not detected for the deaminase domain of hyADAR (hyADARd), the full-length enzyme edited the Gli1 mRNA substate to over 80% completion within 60 minutes with an observed rate of 0.074 ± 0.005 min⁻¹ at 30°C (Figure 4b). Given that Hydra vulgaris survive at temperatures ranging from 18-21°C, we hypothesized hyADAR could have adapted to efficiently perform deamination at lower temperatures. At 18°C the rate of deamination is decreased by approximately 10-fold (0.006 ± 0.003 min⁻¹) and reached an endpoint of less than 50% after 120 minutes (Figure 4b). The observed kinetic rate of human ADAR2 is typically much higher on the human Gli1 mRNA substrate (k_obs > 1 min⁻¹), particularly for the ADAR2 E488Q hyperactive mutant (k_obs > 2 min⁻¹) [19]. However, this study is the first to directly measure deaminase activity by hyADAR, albeit with a human RNA substrate.

Figure 4: Full length hyADAR is an active deaminase.

(a) Deamination activity for full length hyADAR was tested on the well-characterized human substrate Gli1 150mer where the target adenosine is boxed in red [16]. Duplex RNA secondary structure of the Gli1 150mer was predicted using the RNAFold webserver but only the region around the edited Adenosine is shown for simplicity. (b) Kinetic time course under single turnover conditions measuring the rate and endpoint of deamination of the Gli1 duplex RNA substrate by hyADAR measured at 30°C (blue) and 18°C (purple).

High resolution crystal structure of the deaminase domain of hyADAR.

To further characterize hyADAR, we solved the crystal structure of the hyADAR deaminase domain (hyADARd), which is typically easier to crystallize than the modular full-length ADAR. Crystals of hyADARd were grown using the hanging drop vapor diffusion method at room temperature over the course of 10 days. Crystals indexed to the low symmetry triclinic space group P1 so two different data sets from two crystals were merged resulting in a more complete 2.0 Å resolution data set. The structure of hyADARd was solved using molecular replacement with a previous human ADAR2d (PDB ID: 5ED1) [15] structure as a search model and refined to a final R_factor and R_free of 21.1% and 24.3%, respectively (Figure 5, Table 1).

Figure 5: Highly conserved structural features of ADARs are also present in hyADARd.

(a) The 2.0 Å resolution crystal structure of hyADARd showing the catalytic zinc and inositol hexakisphosphate (IP₆) in spheres. The partially resolved 5’ binding loop is colored in light blue, GQG base flipping loop colored in magenta, and the disordered dimerization helix is denoted by a black arrow. (b) The predicted active site of hyADARd is formed around a catalytic zinc coordinated by two cysteines (C628 and C685), a histidine (H574), and a water molecule that’s interacting with the proton shuttle catalytic residue E576. (c) In addition to the presence of IP₆ in the protein core of hyADARd, the highly conserved hydrogen bonding networking linking IP₆ to the active site through K688, D372, and K653 is also observed in hyADARd. Shown in green-colored mesh is the simulated-annealed omit Fo-Fc difference electron density at 3σ, refined with the IP₆ molecule omitted.

Table 1.

Data Processing and Refinement Statistics for hyADAR

PDBID	9dp5
Synchrotron (Beamline)	SSRL 12-2
Wavelength (Å)	0.97946
Space Group	P1
Unit Cell Parameters	a = 49.37Å, b = 49.44Å, c = 83.08Å, α = 83.16°, β = 80.30°, γ = 71.13°
Resolution Range (Å)	62.0 – 2.00 (2.05 – 2.00)
No. observed reflections	165,743 (12,443)
No. unique reflections	47,063 (3,516)
Completeness	95.2% (95.1%)
Redundancy	3.52 (3.54)
I/σ (I)	7.79 (1.23)
Rmergea (%)	9.4 (112.3)
CC1/2 (%)	99.6 (73.7)
Refinement Statistics
Rfactorb (%)	21.1
Rfreeb (%)	24.3
RMS bond length (Å)	0.007
RMS bond angle (°)	1.67
Average B-factor (Å2)	57.4
Ramachandran Plot Statistics c
Favored (%)	91.8
Allowed (%)	98.5
Outliers (%)	1.5
No. of atoms
Protein	5,520
Inositol Hexakisphosphate (IP6)	72 (2 molecules)
Zn	2
Waters	90

^a${\mathrm{R}}_{\text{merge}}=[{\sum }_{\mathrm{h}}{\sum }_{\mathrm{i}}\mid {\mathbf{I}}_{\mathrm{h}}-{\mathbf{I}}_{\text{hi}}\mid ∕{\sum }_{\mathrm{h}}{\sum }_{\mathrm{i}}{\mathbf{I}}_{\text{hi}}]$ where ${\mathbf{I}}_{\mathrm{h}}$ is the mean of ${\mathbf{I}}_{\text{hi}}$ observations of reflection h. Numbers in parenthesis represent highest resolution shell.

^bR-Factor and ^b $\text{Rfree}=\sum \mid \mid {\mathrm{F}}_{\text{obs}}\mid \text{-}\mid {\mathrm{F}}_{\text{calc}}\mid \mid ∕\sum \mid {\mathrm{F}}_{\text{obs}}\mid \mathrm{x}100$ 100 for 95% of recorded data (R-Factor) or 5% data (Rfree).

^cRamachandran plot statistics from MolProbity.

Comparison to previous structures of human ADAR2d revealed that hyADARd is highly structurally similar with an alignment RMSD of 0.652 for 258 equivalent alpha-carbons (PDB ID: 1zy7) [17]. The active site of hyADARd is formed around the catalytic zinc coordinated by a water molecule, H574, C628, C658, and the catalytic residue E576, which hydrogen bonds to the zinc-bound water molecule (Figure 5b). The first structure of human ADAR2d revealed an inositol hexakisphosphate (IP₆) molecule buried in the protein core and demonstrated that the presence of IP₆ is required for proper ADAR folding and deaminase activity [17]. Indeed, strong IP₆ electron density is observed in the core of hyADARd along with the conserved hydrogen bond networking linking the IP₆ to the active site (Figure 5c). Part of the 5’ binding loop of hyADARd is disordered from residue 639-643 and thus is not included in the model, reminiscent of the RNA-free structure of human ADAR2 [17]. Modeling of an RNA duplex (aligned from PDB ID: 5hp3) [15] indicates the hyADARd 5’ binding loop closely approaches the RNA duplex in a similar manner as does human ADAR2 (Supplemental Figure S1) with all but four of the fifteen RNA contacting residues conserved, including the residues directly contacting the flipped out base [15].

While human ADAR2d and hyADARd both form dimers in the asymmetric unit, hyADARd dimerizes differently than human ADAR. The hyADAR dimer (Supplemental Figure S2a) has a buried surface area of 796 Ų. Dimerization is predominately mediated by water molecules between a loop on monomer A and the N-terminus of α-helix 1 of the other hyADAR deaminase domain (Supplemental Figure S2b) with additional dimer interactions not mediated by water found between two loops of the deaminase domains (Supplemental Figure S2c). Given the low buried surface area and water-mediated dimerization, it is likely the hyADARd dimer observed in our structure is a crystal artifact. To determine the native state of hyADARd in solution we performed size exclusion chromatography and compared the retention time of hyADARd to that of known molecular weight standards (Supplemental Figure S2d). From the linear fit of retention time vs. Log MW for known standards, we determined hyADARd eluted as a 40 kDa monomer (Supplemental Figure S2e), suggesting the dimer observed in the structure may not be biologically relevant.

Human ADAR dimerization is mediated by a short α-helix with a highly conserved TWDG motif, where mutation of the motif significantly decreases deamination activity on dimerization dependent substrates [16]. The TWDG dimerization motif is not fully conserved in Hydra vulgaris with the tryptophan being replaced by serine in Hydra and may contribute to the reason hyADAR does not dimerize like human ADARs. Much like the 5’ binding loop length, the conservation of the TWDG dimerization motif diverges in the Cnidaria phylum with clade 1 of the anthozoan subphylum retaining the TWDG motif, while clade 2 has a TIDG motif. In the medusozoan subphylum, position two of the dimerization motif is encoded by either serine, methionine, or threonine with Hydra species exhibiting a TSDG motif (Figure 6a). In our hyADARd structure, residues 665-678, which contains the ^670-TSDG⁻⁶⁷³ motif, are disordered potentially due to the loss of tryptophan within the dimerization motif. In the structure of human ADAR2-R2D (PDB ID: 6vff), W502 of the TWDG motif on the RNA-engaged catalytic monomer contacts the backbone carbonyl of F547 of the RNA-disengaged auxiliary monomer (Figure 6b), raising question if hyADAR could dimerize in a similar manner to human ADARs in the presence of RNA.

Figure 6: Medusozoan cnidarians lack conservation of TWDG dimerization helix.

(a) While humans, several model organisms, and anthozoan cnidarians have a conserved TWDG motif that forms a short helix required for asymmetric dimerization, medusozoan cnidarians including Hydras lack complete conservation of this motif. (b) Human ADAR2 dimer interface highlighting interactions mediated by the dimer helix containing the TWDG motif (PDB ID: 6vff).

Discussion

While the role of ADAR editing in human biology and disease has and continues to be extensively documented [6, 33] studies investigating ADAR editing in evolutionary distant organisms has been limited to studies of C. elegans and cephalopod ADARs [34-38]. Further characterization of ADAR editing in a diverse range of metazoans may uncover unique biological consequences and regulation of RNA editing as well as provide a more comprehensive understanding of ADAR evolution.

In this study, we mapped the evolutionary relationships of ADARs both within and beyond the cnidarian phylum. Using phylogenetic analyses, we found ADARs from the cnidarian subphylum Medusozoa formed a single clade that clustered with other ADAR2 genes. In contrast, the ADAR2s from the anthozoan subphylum formed two distinct clades with Clade 1 resembling non-cnidarian ADAR2s and Clade 2 closely related to ADAR1 genes. AlphaFold3 predictions of cnidarian ADARs domain architecture indicated the majority of medusozoan and Clade 2 anthozoan ADARs contained five dsRBDs with Clade 1 anthozoan ADARs predicted to have only two dsRBDs (Figure 1). AlphaFold3 predicted hyADAR dsRBD1-5 structures showed high structural similarity with the structure of dsRBD2 from human ADAR2 with many residues important for domain architecture and RNA binding conserved (Figure 2). Multiple sequence alignment of the deaminase domain of hyADAR with other cnidarian ADARs showed an association with shorter 5’ binding loops and a glutamine base flipping residue in the medusozoan subphylum (Figure 3). Adenosine-to-inosine kinetic assays of hyADAR with a model human substrate confirmed it is an active deaminase (Figure 4). The crystal structure of hyADARd revealed conserved active site architecture and the presence of IP₆ essential for ADAR activity. The short 5’ binding loop of hyADAR was partially resolved and adopted a structure similar to the 5’ binding loop observed in ADAR2 complexed with RNA (Figure 5, Supplementary Figure S1). Multiple sequence alignment of ADARs within the Cnidaria phylum indicate medusozoan ADARs lack full conservation of a TWDG motif important for dimerization (Figure 6). Indeed, the structure of hyADAR forms a dimer dissimilar to human ADAR2 with a dimer interface predominantly mediated by water molecules.

Although the amino acid recoding potential of ADARs is attractive for the development of site-directed RNA editing therapeutics, natural recoding events are relatively rare with less than 3% of human transcripts containing ADAR edited recoding sites, 1-4% detected in Drosophila, and only a few recoding sites documented in C. elegans, with many recoding events lacking functional significance [39, 40] Instead, the vast majority of A-to-I RNA editing sites occur in repeated mobile elements like Alu retrotransposons and may contribute to the maintenance of genomic stability and integrity and dampen the innate immunity in dsRNA detection [41, 42]. Editing sites not occurring in mobile genetic elements typically occur in intronic and intergenic regions, contributing to ADARs role in regulating gene expression, splicing, and RNA structure [43, 44]. Conversely, the transcriptome of some ectotherms such as the cephalopods squid (Doryteuthis paeleii), octopus (Octopus vulgaris and Octopus bimaculoides) and cuttlefish (Sepia oficianalis) appear to be highly edited with recoding sites enriched in genes with neuronal function [45]. Furthermore, Garret et al. demonstrated the role of ADAR editing in temperature adaption by comparing potassium channel orthologs from tropical and Antarctic octopi, who live at 30°C and −1.8°C, respectively. They hypothesized that mutations in the potassium channel gene would have accumulated in the Antarctic octopus to allow for efficient action potential firing at colder temperatures. However, while the two channels were functionally equivalent at their respective native temperatures, their amino acid sequence only differed at four positions, suggesting an alternative mechanism of adaptation beyond genetic natural selection. Indeed, five editing sites were found to be species specific and resulted in amino acid changes that altered action potential kinetics [46]. Given that Hydra, and cnidarians more broadly, live at various temperatures, ADAR editing may contribute to temperature adaptation at the transcriptome level.

Lastly, while spatiotemporal studies of RNA editing in humans has revealed highly dynamic landscapes of A-to-I editing in several different tissues, especially neural tissues [47-50], only a few studies have assessed the spatiotemporal nature of ADAR editing in lower complexity metazoans. One such study of the cnidarian coral Acropora millepora revealed increased RNA editing levels in non-coding regions during a spawning event and high editing levels detected in newly released gametes, potentially introducing variability in the coral gametes [51]. ADAR RNA editing could serve as a general function in cnidarians, including Hydra, to diversify offspring during sexual reproduction.

Given the lack of structural data of non-human ADARs, we are limited in our comparison of the structure of hyADARd to other metazoan ADARs. While hyADAR is structurally very similar to human ADAR2 with conserved active site architecture and RNA binding residues along with the presence of the buried IP₆, they vastly differ in their oligomeric structure. Human ADARs have been structurally and biochemically shown to dimerize, with one deaminase monomer directly interacting with the RNA and flipping out the target adenosine, while the deaminase domain of the second monomer does not make any RNA contacts [15, 17]. Recent structures of truncated human ADAR2 with the deaminase domain and dsRBD2 show the dsRBD2 of the RNA disengaged deaminase “auxiliary” monomer interacting with the duplex in the 3’ direction relative to the flipped-out base [16, 21]. While human ADAR2 dimerizes in the absence of dsRNA [17], it is possible that hyADAR adopts the human-like dimeric form when bound to duplex RNA, which can be explored in future studies by solving structures for the full length hyADAR bound to highly edited native Hydra vulgaris substrates.

While high structural similarity of hyADARd and human ADAR2d implies similar enzymatic activity, the differing oligomeric structure, base-flipping residue, additional dsRBDs, together with the shorter 5’ binding loop, may suggest dissimilar RNA substrates potentially resulting in distinctive biological phenotypes. Taken together, studies of evolutionary diverse organisms suggest that ADARs have evolved specific functions for specific environments and rigorous characterization of non-human ADARs is necessary to uncover unique biological roles.

Experimental Procedures

Generation of ADAR phylogenetic tree

Candidate ADAR1, ADAR2 and ADAT1 sequences were identified from published cnidarian genomes [52-62] by BLAST search [63] with BLAST+ [64]. Reverse BLAST to the National Center for Biotechnology (NCBI) non-redundant protein sequence (nr) database [23] was used to further confirm identity, in addition to examining conserved domains for select sequences. Model organism ADAR1, ADAR2 and ADAT1 sequences were also collected from the NCBI nr database [23]. A multiple sequence alignment of these sequences was performed with MAFFT v7.526 [65] using the 'auto' option. This alignment was trimmed of evolutionarily uninformative regions with ClipKIT (browser settings 'Mode: smart-gap', 'Sequence-Type: Amino acids') [66]. IQ-TREE 2 v2.3.6 [67] was used to infer a maximum likelihood gene tree from the trimmed alignment using the following command: `qtree2 -s alignment.fasta -m MFP -st CODON -B 1000 -T AUTO --seqtype A`. ModelFinder [68], as implemented in IQ-TREE 2, was employed to identify the best-fitting evolutionary model, which was the Q.pfam+F+I+R5 model based on the Bayesian Information Criterion. Support scores were calculated using 1000 bootstrap replicates. The gene tree was rooted with the ADAT1 clade, which was chosen as the outgroup as ADARs likely evolved from an ADAT1-like sequence [29, 69]. The final figure was refined using the `ape` package (v5.8) [70] in R (v4.4.2) (https://www.R-project.org/) [71], FigTree v1.4.4 (http://tree.bio.ed.ac.uk/software/figtree/), and Adobe Illustrator 2024. For visualization, the ADAT1 outgroup clade was collapsed.

Purification of full-length Hydra vulgaris ADAR for kinetic assays.

The gene encoding for full-length ADAR from Hydra vulgaris (hyADAR) or the deaminase domain (hyADARd) were purchased as g-blocks (Integrated DNA technologies) and cloned into the pSc-ADAR yeast expression plasmid using Gibson Assembly cloning. Overexpression of hyADAR and hyADARd were conducted in Saccharomyces cerevisiae as previously described for human ADAR [15, 72]. Cells containing overexpressed protein were lysed using a microfluidizer in cold lysis buffer containing 20 mM Tris-HCl pH 8.0, 750 mM NaCl, 35 mM imidazole, 5% glycerol, 0.01% TritonX-100, 1 mM β-mercaptoethanol, and 1 Roche cOmplete^™ mini EDTA-free protease inhibitor tablet. Lysate was clarified by centrifugation at 39,000 x g for 45 min and soluble lysate was filtered through a 0.45 μm PES (Polyethersulfone) filter to remove the residual lipid layer and cell debris. Clarified lysate was then passed over a 1 mL Ni-NTA column and washed with 20 mL lysis buffer, 60 mL wash buffer I (20 mM Tris-HCl pH 8.0, 400 mM NaCl, 35 mM imidazole, 5% glycerol, 0.01% Triton x 100, 1 mM β-mercaptoethanol), and 30 mL wash buffer II (20 mM Tris-HCl pH 8.0, 100 mM NaCl, 35 mM imidazole, 5% glycerol, 0.01% Triton x 100, 1 mM β-mercaptoethanol). HyADAR was collected by a 60 mL linear gradient elution in wash buffer II and elution buffer (20 mM Tris-HCl pH 8.0, 100 mM NaCl, 300 mM imidazole, 5% glycerol, 0.01% Triton x 100, 1 mM β-mercaptoethanol). Fractions corresponding to eluted hyADAR were pooled and then loaded onto a 1 mL Hi-Trap Heparin HP column at a flow rate of 0.5 mL/min then washed with 30 mL wash buffer II and eluted in a 60 mL linear gradient of wash buffer II and heparin elution buffer (20 mM Tris-HCl pH 8.0, 1M NaCl, 5% glycerol, 1 mM β-mercaptoethanol). Protein concentration was determined using BSA standard densitometry and aliquots were frozen in storage buffer (20 mM Tris pH 8.0, 200 mM NaCl, 20% glycerol, 1 mM β-mercaptoethanol) at −80°C.

Preparation of Gli1 150nt RNA substrate for in vitro kinetics.

The Gli1 150nt self-folding substrate was prepared from linearized DNA using the HiScribe^™ T7 High Yield RNA Synthesis Kit (New England Biosciences) [16]. Single-stranded RNA was purified by denaturing polyacrylamide gel electrophoresis and the target band excised, crushed, and soaked in 0.5 M NaOAc, 0.1% SDS, and 0.1 mM EDTA overnight at 4°C. Solubilized RNA was collected by passing the polyacrylamide gel slurry through a 0.2 μm filter and collected by ethanol precipitation and lyophilized to dryness. Purified RNA was self-folded in a solution containing 180 nM Gli1 150 RNA, 1X TE buffer, and 100 mM NaCl and heated to 95°C for 5 min then slow cooled to room temperature.

In vitro deamination kinetics of hyADAR.

In vitro deamination kinetics were conducted under single-turnover conditions. Reaction mixtures containing 15 mM Tris-HCl pH 7.5, 3% glycerol, 60 mM KCl, 1.5 mM EDTA, 0.003% Nonidet P-40, 3 mM MgCl₂, 160 U/mL RNAsin, 1.0 μg/mL yeast tRNA, 10 nM Gli1 150 folded RNA were incubated at 30°C or 18°C for 30 min before the addition of enzyme to a final concentration of 100 nM. At time points corresponding to 0, 0.5, 5, 15, 30, 60, 90, and 120 min the reaction was quenched by addition of 190 μL 95°C nuclease-free water, vortexed, and heated to 95°C for 5 min. RNA from quenched reactions was amplified using the Promega Access RT-PCR system to generate cDNA which was then purified and subjected to Sanger Sequencing. The peak height of adenosine and guanosine form the sequencing chromatogram were quantified using 4Peak v1.8. Each experiment was conducted in triplicate and the percent editing was calculated using the equation: $\%\mathit{editing}=\frac{\mathit{G}\mathit{Peak}\mathit{Height}}{\mathit{G}\mathit{Peak}\mathit{Height}+\mathit{A}\mathit{Peak}\mathit{Height}}\times 100$. Data were plotted in Prism (GraphPad) and fit the equation: ${[\mathit{P}]}_{\mathit{t}}={\mathit{P}}_{\mathit{f}}\times (1-{\mathit{e}}^{-{\mathit{k}}_{\mathit{obs}}(\mathit{t})})$ where ${[\mathit{P}]}_{\mathit{t}}$ is the percent editing at time t, ${\mathit{P}}_{\mathit{f}}$ is the end point of editing, and ${\mathit{k}}_{\mathit{obs}}$ is the observed rate constant.

Purification of Hydra vulgaris ADAR deaminase domain (hyADARd) for X-ray crystallography studies.

hyADARd was overexpressed and purified similar to hyADAR with the following exceptions. Following Ni-NTA purification, hyADARd was not purified by heparin affinity chromatography because it does not bind to the heparin column. Immediately succeeding Ni-NTA purification, the 10xHis tag of hyADARd was cleaved in an optimized 1:1 mass ratio of His-tagged TEV protease at room temperature for 2 hr without agitation. Cleaved hyADARd was collected by passing the cleavage reaction through a 1 mL Ni-NTA column at a flow rate of 0.5 mL/min followed by a TEV Ni-NTA wash buffer containing 20 mM Tris pH 8.0, 300 mM NaCl until hyADARd flowing through the column could no longer be detected by the absorbance at 280 nm. Fractions corresponding to 10xHis tag cleaved hyADARd were pooled and dialyzed against size exclusion buffer containing 20 mM Tris pH 8.0, 200 mM NaCl, 5% glycerol, 1 mM β-mercaptoethanol followed by size exclusion chromatography using a HiLoad 16/600 Superdex PG column. Size exclusion eluted fractions corresponding to hyADARd were pool and concentrated to 4.2 mg/mL (100 μM) for crystallization experiments.

Size -exclusion chromatography.

Hydra ADARd was concentrated to 1 mL, filtered through a 0.2 μM filter, and injected onto a GE Healthcare HiLoad 16/600 Superdex PG column with 120 ml column volume at a flow rate of 0.5 ml/min. The elution volume (Ve) decreases nearly linear with the log of the molecular hydrodynamic volume. For calibrating the column, a mixture of protein standards (Bio-Rad) consisting of thyroglobin (670 kDa), γ-globulin (158 kDa), ovalbumin (44 kDa), myoglobin (17 kDa) and vitamin B12 (1.35 kDa) was used. The calibration standard thyroglobulin (670 kD) was used to determine the void volume (Vo) of the column. Analysis is carried out by dividing the Ve of the standards by the Ve of the thyroglobulin (Ve/Vo) and plotting against the log of the MW of the standard.

Crystallization of hyADARd.

Crystals of hyADARd were grown using the hanging-drop vapor diffusion method at room temperature. A 150 nL solution of 4.2 mg/mL hyADARd in size exclusion buffer was mixed with equal volume of mother liquor containing 0.2 M NaCl, 0.1 M Bis-Tris pH 6.5, 25% PEG3350. Following 10-day equilibration, small rod-shaped crystals formed, and a single rod was harvested following a soak in mother liquor containing 30% glycerol before plunge freezing in liquid nitrogen.

X-Ray Diffraction Data Processing, Structure Determination and Refinement.

Diffraction data of two crystals were collected using the fine-phi slicing method at SSRL beamline 12-2 and were processed using XDS [73], scaled and merged using the Aimless program in the CCP4 data processing package [74]. Due to the low symmetry P1 space group, diffraction data from two crystals were merged to produce a more complete (>95%) data set with an overall R_merge of 9.4%. The Matthews coefficient calculated using CCP4 [75] indicated two monomers in the asymmetric unit (unit cell of space group P1). The structure of hyADARd was solved by molecular replacement using the previous structure of a mutant human ADARd (PDB ID: 5ed1)[15] as a model in the Phaser program of the CCP4 data processing package [74]. The dsRNA was removed from the search model and non-conserved protein residues were truncated to alanine using the CHAINSAW program in the CCP4 data processing package [74, 76]. Next, the phaser output model was further built using Autobuild and refinement in PHENIX [77] was conducted using TLS parameters, simulated annealing, and Zn coordination restraints. Model building and adjustment was completed using COOT [78]. Ideal Zn coordination distances were determined using the MetalPDB database as well as Zn-ligand distances observed in previous ADAR structures[79]. Final data processing and refinement statistics are given in Table 1. Atomic coordinates and Structure Factors were deposited in the RCSB protein data bank with PDB ID 9dp5.

Highlights.

• Phylogenetic and structural analysis of cnidarian ADARs suggests they have five double-stranded RNA binding domains with conserved amino acids known to be important for RNA binding.

• Evolutionary divergence amongst ADARs in the anthozoan and medusozoan subphyla of cnidarians with medusozoan ADARs (including Hydra) having short 5’ RNA binding loops, glutamine flipping residues, and loss of conservation of an important dimerization motif.

• First structure of a non-human ADAR deaminase domain.